基于多源数据的特长隧道驾驶疲劳模型  

A Driving Fatigue Model for Extra-long Tunnels Based on Multi-source Data

作　　者：尚婷[1] 连冠 黄龙显 谢磊 SHANG Ting;LIAN Guan;HUANG Xianlong;XIE Lei(School of Traffic&Transportation,Chongqing Jiaotong University,Chongqing 400074,China;School of Civil Engineering,Chongqing Jiaotong University,Chongqing 400074,China;Chongqing High Speed Engineering Consulting Co.,Ltd.,Chongqing 404100,China)

机构地区：[1]重庆交通大学交通运输学院,重庆400074 [2]重庆交通大学土木工程学院,重庆400074 [3]重庆高速工程顾问有限公司,重庆404100

出　　处：《交通信息与安全》2024年第4期30-41,共12页Journal of Transport Information and Safety

基　　金：国家自然科学基金项目(52172341);教育部青年人文社会科学研究青年基金项目(22YJCZH143);重庆市交通委员会交通科技项目(CQJT2022ZC06)资助。

摘　　要：为研究驾驶人在特长隧道内驾驶疲劳演变过程及其影响因素,基于实车试验采集的多源数据,对特长隧道内驾驶疲劳分类判别以及驾驶疲劳影响因素关系模型展开了研究。通过差异显著性分析和相关性分析筛选出闭眼百分率P80、瞳孔直径变异系数和加速度作为疲劳敏感性指标,并分析了各指标随行驶时间累积的变化规律。为构建驾驶疲劳分类判别模型,基于卡罗林斯卡嗜睡量表(Karolinska sleeping scale,KSS)主观疲劳检测结果,将疲劳程度划分清醒状态、半疲劳状态和疲劳状态,采用构造多类分类器的方法将不同疲劳状态样本进行组合分类,利用网格搜索法进行分类模型的参数寻优,并将筛选出的疲劳敏感性指标作为分类模型的输入变量,建立了基于网格搜索法的多分类支持向量机疲劳状态判别模型(GS-M-SVMs模型)。然后根据疲劳状态分类判别模型,利用有序多分类Logistic模型建立了特长隧道疲劳程度与影响因素的关系模型,对特长隧道内驾驶疲劳影响因素进行了探究。研究结果表明:疲劳敏感性指标变化规律可有效表征特长隧道内驾驶疲劳演变过程,而GS-M-SVMs模型分类检测准确率达到90.75%,对疲劳程度的分类识别效果较好,并且累积行驶时间和隧道长度显著影响驾驶人的疲劳程度,其模型回归系数分别为2.634和0.395,表明累积行驶时间是驾驶人在特长隧道路段中疲劳程度加重的最主要因素,隧道照度和隧道线形等因素并无显著影响。To investigate the evolution of driving fatigue in extra-long tunnels and its influencing factors,multi-source data from real-vehicle experiments are utilized to classify and identify driving fatigue,as well as to ana-lyze the relationship between fatigue levels and influencing factors.Through significance tests of differences and correlation analysis,the percentage of eyelid closure over the pupil over time(PERCLOS)P80,the variable coeffi-cient of pupil diameter,and acceleration are selected as key fatigue sensitivity indicators,and their changing pat-terns with accumulated driving time are examined.To construct a driving fatigue classification model,fatigue lev-els,based on the subjective fatigue detection results from the Karolinska sleepiness scale(KSS),are categorized in-to awake,semi-fatigued,and fatigued states.A multi-class classifier method is then employed to combine and classi-fy these fatigue states.The grid-search method(GS)is utilized for parameter optimization,and the selected fatigue sensitivity indicators are used as input variables to establish a multi-class support vector machine model(GS-M-SVMs)for fatigue state classification.Following this,an ordinal multi-class Logistic model is developed to explore the relationship between driving fatigue levels and influencing factors in extra-long tunnels.The results indi-cate that the changing patterns of fatigue sensitivity indicators effectively capture the evolution of driving fatigue.The GS-M-SVMs model achieved a classification accuracy of 90.75%,indicating strong performance in fatigue lev-el detection.Both accumulated driving time and tunnel length significantly influence driving fatigue levels,with re-gression coefficients of 2.634 and 0.395,respectively.This indicates that accumulated driving time is the primary factor contributing to increased fatigue in extra-long tunnels,while factors such as tunnel illumination and align-ment do not significantly impact fatigue levels.

关 键 词：交通安全 驾驶疲劳 GS-M-SVMs模型 网格搜索法 有序多分类Logistic模型 

分 类 号：U491[交通运输工程—交通运输规划与管理]